MRF-UNets: Searching UNet with Markov Random Fields

TL;DR

This paper introduces MRF-NAS, a neural architecture search framework based on Markov Random Fields, to optimize UNet architectures for semantic segmentation, leading to significant performance improvements across multiple datasets.

Contribution

The paper proposes MRF-NAS, a novel NAS method utilizing MRFs for architecture exploration, and designs MRF-UNet, an optimized UNet variant with superior segmentation performance.

Findings

MRF-UNet outperforms benchmarks on aerial and medical datasets

The approach efficiently explores architectures with loopy inference graphs

MRF-UNetV2 further improves results over the original MRF-UNet

Abstract

UNet [27] is widely used in semantic segmentation due to its simplicity and effectiveness. However, its manually-designed architecture is applied to a large number of problem settings, either with no architecture optimizations, or with manual tuning, which is time consuming and can be sub-optimal. In this work, firstly, we propose Markov Random Field Neural Architecture Search (MRF-NAS) that extends and improves the recent Adaptive and Optimal Network Width Search (AOWS) method [4] with (i) a more general MRF framework (ii) diverse M-best loopy inference (iii) differentiable parameter learning. This provides the necessary NAS framework to efficiently explore network architectures that induce loopy inference graphs, including loops that arise from skip connections. With UNet as the backbone, we find an architecture, MRF-UNet, that shows several interesting characteristics. Secondly, through the lens of these characteristics, we identify the sub-optimality of the original UNet architecture and further improve our results with MRF-UNetV2. Experiments show that our MRF-UNets significantly outperform several benchmarks on three aerial image datasets and two medical image datasets while maintaining low computational costs. The code is available at: https://github.com/zifuwanggg/MRF-UNets.